O. trifallax is a unicellular eukaryote harbouring two kinds of nuclei: germline micronuclei and somatic macronuclei. Diploid micronuclei are transcriptionally inert during vegetative growth but they transmit the germline genome through subsequent generations. Effectively polyploid macronuclei provide all vegetative gene expression, but degrade after fertilization, when new micronuclei and macronuclei develop. DNA differentiation in ciliates such as Oxytricha (also called Sterkiella) involves massive chromosome fragmentation and deletion of transposons and internally eliminated sequences (IESs), accomplishing 95% genome reduction, compressing a 1 Gb germline into one-twentieth of the space. Rearrangement of the remaining short segments (macronuclear destined segments; MDSs) by permutation or inversion, followed by telomere addition and amplification, produces mature macronuclear ‘nanochromosomes’, typically ~2 kb with just one gene¹².

Genome-wide rearrangements discard nearly all nongenic DNA in Oxytricha, packing a streamlined gene-rich (~30,000 genes) eukaryotic genome in only 50 Mb. Furthermore, hundreds of thousands of MDSs require unscrambling (reordering or inversion) to interpret the sequence information in their DNA. The remarkable degree of specificity and reproducibility suggests a highly accurate mechanism to programme rearrangements. The parental cell must provide sufficient information to assemble a fully functional macronucleus. The mechanisms that allow perfect recognition of hundreds of thousands of DNA sequences for elimination or reordering remain largely unknown. The discovery of homology-dependent maternal effects that modify DNA deletions may provide a clue^13,14.

Homology-dependent trans-nuclear comparison of germline and somatic genomes may regulate these DNA rearrangements. In Tetrahymena, germ-line-specific small RNAs called scan RNAs derive from the micronuclear-limited sequence and target DNA for elimination¹⁵. However, this model cannot explain fusion of unlinked or disordered DNA segments in Oxytricha, which can join segments smaller than the length of typical scan RNAs. DNA rearrangements in Tetrahymena are imprecise, involving no permutations.

In other species including Paramecium, DNA rearrangements occur at short direct repeats called pointers. Homologous recombination between identical repeats at MDS–IES boundaries can both remove the IESs between them and reorder MDSs, leaving one copy of the pointer in the final product. Pointers of 2–20 bp in Oxytricha (average length of 4 bp between nonscrambled junctions and 9 bp between scrambled junctions¹⁶) generally occur at the 3′ end of every MDS segment n and the 5′ end of segment n + 1, providing a simple linked list of final MDS order. Although necessary for DNA deletion in Paramecium¹⁷, 2–20 bp pointers could not unambiguously guide rearrangement¹⁸.

RNA interference against putative RNA templates leads to aberrant (incorrect or completely blocked) gene unscrambling in the resulting progeny (Fig. 1). PCR screening of TEBPα segments 5–17, after treatment with the RNAi vector spanning segments 2–17 in order, reveals DNA molecules longer than the typical macronuclear product, but shorter than the micronuclear precursor (Fig. 1a, lane 1). Sequencing revealed these were partially and/or incorrectly rearranged. Some aberrant forms still contain IESs (Fig. 1b, examples 2–5 and 9), with a bias for retention between scrambled segments (black boxes in Fig. 1). IESs between nonscrambled segments (red in Fig. 1) are usually absent. This suggests different susceptibility between these two kinds of IESs to RNAi treatment, and possibly a mechanistic decoupling, consistent with observations that simple IES removal precedes unscrambling in the actin I gene in the related genus Stylonychia²³. Some permutations (partial descrambling) occurred (Fig. 1b, examples 4–6; molecule 4 could be an unscrambling intermediate), and even large duplications (Fig. 1b, examples 7 and 8) or deletions (Fig. 1b, examples 8–9). All deletions occur between short direct repeats that provide alternative (cryptic) pointers for recombination—the type of error that occurs early during rearrangement²³.

RNAi against pol-α leads to almost complete abrogation of DNA elimination or rearrangement (Fig. 1c, lane 2) with no permutations at all, but with several examples of aberrant deletion at cryptic pointers (Fig. 1d, examples 3–7), including one large deletion between segments 4 and 31 (Fig. 1d, example 7). Approximately 85% of pol-α molecules surveyed between segments 3 and 31 (the RNAi vector spanned segments 16–29) 80 h post conjugation were identical in size to the germline precursor (Fig. 1c, lane 2, and Fig. 1d, example 2), indicating no rearrangement (which we confirmed by sequencing).

We designed artificial macronuclear chromosomes to permute the natural order of two DNA segments in both TEBPα (Fig. 3a) and TEBPβ (Fig. 3c). TEBPα is scrambled in the germ line, requiring several DNA permutations to assemble a functional gene during development. TEBPβ normally requires only DNA deletion and no permutation (with identical MDS order in both nuclei). We chose these genes to test whether template-directed rearrangements act genome-wide or are specific to scrambled genes.

In two independent experiments, microinjecting the double-stranded, synthetic TEBPα chromosome (labelled ‘TEBPα sw78’ template in Fig. 3, where SW indicates switched segment order) into the macronucleus of conjugating pairs of Oxytricha switches the order of segments 7 and 8, and injecting the synthetic TEBPβ molecule (‘TEBPβ sw45’ template) switches the order of segments 4 and 5. Each template slightly adjusts segment boundaries (Supplementary Fig. 1) to recruit existing sequence repeats as alternative pointers at new recombination junctions. Both templates contain single nucleotide substitutions (some creating restriction sites) to distinguish microinjected DNA from processed endogenous genes. Non-injected cells were used as controls. DNA was extracted from progeny one week after microinjection, permitting sufficient asexual growth. Injected cells grew slower than non-injected cells. In both cases, the F₁ progeny contained alternatively rearranged macronuclear chromosomes, following the reprogrammed order (TEBPα sw78 and TEBPβ sw45; Fig. 3b, d, e). In one case, the switched order was even the major product (Fig. 3d, lane 3).

Restriction mapping ruled out either the presence of microinjected DNA in the harvested cells or extended copying or incorporation of templates during rearrangement (Fig. 3b, lanes 5–8, and Fig. 3d, lanes 17–22). (Sequence analysis revealed a few exceptions; see section entitled ‘Transfer of nucleotide substitutions’.) For TEBPα, the new junction formed between segments 6 and 8 creates a Tsp509I site, which distinguishes wild-type from switched chromosomes (Fig. 3b, lanes 1–4). For TEBPβ, the presence of two restriction sites (BbvCI and BsrGI) in MDS4 produces differently sized products in switched versus wild-type molecules (Fig. 3d, lanes 10–15). In addition, for TEBPβ, a simple PCR assay measuring the length from segment 4 to 6 yields a smaller product for switched versus wild-type chromosomes, because MDS5 is missing in the switched PCR product (Fig. 3d, lanes 2–8). PCR screening was not feasible for TEBPα because segments 7 and 8 are very small, encoding 6 and 10 amino acids, respectively, within the same α helix of the TEBPα protein²⁴.

We also examined DNA from TEBPβ sw45-injected cells two weeks post injection (Fig. 3d, lanes 4–5). These cells have an increase in the ratio of wild-type to switched chromosomes, possibly owing to a fitness decrease in cells burdened with the altered TEBPβ gene. Premature stop codons in the permuted version could produce a severely altered protein or possibly invoke nonsense-mediated mRNA decay (whereas the TEBPα permutation would cause a modest change in helix B of the encoded protein). Thus, we suggest that the use of synthetic templates to reprogramme DNA rearrangement provides a convenient epigenetic tool for reverse genetics in Oxytricha.

To test whether this epigenetic effect transferred to sexual offspring, we induced sexual conjugation by starving TEBPβ sw45-injected cells after two weeks of vegetative growth. Conjugating pairs were isolated and grown separately. Asexual progeny of individual pairs are denoted sw-A and sw-B. We detected the alternative rearrangement pattern in putative F₂ progeny (Fig. 3d, lanes 6–8, and Fig. 3e) and even in some putative F₃ (Supplementary Fig. 2), suggesting stable epigenetic inheritance of this alternative form. In one case (F₂ sw-B) the newly programmed pattern still predominates over wild type.

Our results are compatible with an earlier result showing that RNA injection in Tetrahymena leads to deletion of homologous sequences during development²⁵. The injected RNA might trigger destruction of homologous RNA templates in Tetrahymena, for example, as in our RNAi experiments.

The transfer of some mutations from the template to the rearranged molecule (Supplementary Information) suggests that RNA could be a template for DNA synthesis or repair very close to the pointer, as in template-guided DNA repair⁹. A C-to-T mutation in the template that creates a TA dinucleotide close to a new pointer transferred efficiently to all rearranged molecules, whereas substitutions further away from a pointer transferred infrequently, consistent with local polymerase activity. The RNA-guided DNA rearrangements proposed here are distinct from previous indirect roles for RNA in mediating recombination^26–30.